Data analysis of discrete data using principal/independent component methods. The discrete data can take the form of counts normalized in different ways (e.g., Poisson data, multinomial data, a set of different multinomial data, etc). Classical PCA makes hidden Gaussian assumptions that are a poor fit for low-count discrete data better modelled using Poisson analysis. ICA is also targetted mostly at the same kind of data as PCA. The methods are also a form of ICA, and tight relationships exist between all the different variations, GOM, NMF, PLSI, PLSA, LDA, GaP, MPCA, etc.